Loading...
1/*
2 * Copyright (C) 2001 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include "dm.h"
9
10#include <linux/module.h>
11#include <linux/vmalloc.h>
12#include <linux/blkdev.h>
13#include <linux/namei.h>
14#include <linux/ctype.h>
15#include <linux/string.h>
16#include <linux/slab.h>
17#include <linux/interrupt.h>
18#include <linux/mutex.h>
19#include <linux/delay.h>
20#include <linux/atomic.h>
21
22#define DM_MSG_PREFIX "table"
23
24#define MAX_DEPTH 16
25#define NODE_SIZE L1_CACHE_BYTES
26#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
27#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
28
29/*
30 * The table has always exactly one reference from either mapped_device->map
31 * or hash_cell->new_map. This reference is not counted in table->holders.
32 * A pair of dm_create_table/dm_destroy_table functions is used for table
33 * creation/destruction.
34 *
35 * Temporary references from the other code increase table->holders. A pair
36 * of dm_table_get/dm_table_put functions is used to manipulate it.
37 *
38 * When the table is about to be destroyed, we wait for table->holders to
39 * drop to zero.
40 */
41
42struct dm_table {
43 struct mapped_device *md;
44 atomic_t holders;
45 unsigned type;
46
47 /* btree table */
48 unsigned int depth;
49 unsigned int counts[MAX_DEPTH]; /* in nodes */
50 sector_t *index[MAX_DEPTH];
51
52 unsigned int num_targets;
53 unsigned int num_allocated;
54 sector_t *highs;
55 struct dm_target *targets;
56
57 unsigned integrity_supported:1;
58
59 /*
60 * Indicates the rw permissions for the new logical
61 * device. This should be a combination of FMODE_READ
62 * and FMODE_WRITE.
63 */
64 fmode_t mode;
65
66 /* a list of devices used by this table */
67 struct list_head devices;
68
69 /* events get handed up using this callback */
70 void (*event_fn)(void *);
71 void *event_context;
72
73 struct dm_md_mempools *mempools;
74
75 struct list_head target_callbacks;
76};
77
78/*
79 * Similar to ceiling(log_size(n))
80 */
81static unsigned int int_log(unsigned int n, unsigned int base)
82{
83 int result = 0;
84
85 while (n > 1) {
86 n = dm_div_up(n, base);
87 result++;
88 }
89
90 return result;
91}
92
93/*
94 * Calculate the index of the child node of the n'th node k'th key.
95 */
96static inline unsigned int get_child(unsigned int n, unsigned int k)
97{
98 return (n * CHILDREN_PER_NODE) + k;
99}
100
101/*
102 * Return the n'th node of level l from table t.
103 */
104static inline sector_t *get_node(struct dm_table *t,
105 unsigned int l, unsigned int n)
106{
107 return t->index[l] + (n * KEYS_PER_NODE);
108}
109
110/*
111 * Return the highest key that you could lookup from the n'th
112 * node on level l of the btree.
113 */
114static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
115{
116 for (; l < t->depth - 1; l++)
117 n = get_child(n, CHILDREN_PER_NODE - 1);
118
119 if (n >= t->counts[l])
120 return (sector_t) - 1;
121
122 return get_node(t, l, n)[KEYS_PER_NODE - 1];
123}
124
125/*
126 * Fills in a level of the btree based on the highs of the level
127 * below it.
128 */
129static int setup_btree_index(unsigned int l, struct dm_table *t)
130{
131 unsigned int n, k;
132 sector_t *node;
133
134 for (n = 0U; n < t->counts[l]; n++) {
135 node = get_node(t, l, n);
136
137 for (k = 0U; k < KEYS_PER_NODE; k++)
138 node[k] = high(t, l + 1, get_child(n, k));
139 }
140
141 return 0;
142}
143
144void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
145{
146 unsigned long size;
147 void *addr;
148
149 /*
150 * Check that we're not going to overflow.
151 */
152 if (nmemb > (ULONG_MAX / elem_size))
153 return NULL;
154
155 size = nmemb * elem_size;
156 addr = vzalloc(size);
157
158 return addr;
159}
160EXPORT_SYMBOL(dm_vcalloc);
161
162/*
163 * highs, and targets are managed as dynamic arrays during a
164 * table load.
165 */
166static int alloc_targets(struct dm_table *t, unsigned int num)
167{
168 sector_t *n_highs;
169 struct dm_target *n_targets;
170 int n = t->num_targets;
171
172 /*
173 * Allocate both the target array and offset array at once.
174 * Append an empty entry to catch sectors beyond the end of
175 * the device.
176 */
177 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
178 sizeof(sector_t));
179 if (!n_highs)
180 return -ENOMEM;
181
182 n_targets = (struct dm_target *) (n_highs + num);
183
184 if (n) {
185 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
186 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
187 }
188
189 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
190 vfree(t->highs);
191
192 t->num_allocated = num;
193 t->highs = n_highs;
194 t->targets = n_targets;
195
196 return 0;
197}
198
199int dm_table_create(struct dm_table **result, fmode_t mode,
200 unsigned num_targets, struct mapped_device *md)
201{
202 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
203
204 if (!t)
205 return -ENOMEM;
206
207 INIT_LIST_HEAD(&t->devices);
208 INIT_LIST_HEAD(&t->target_callbacks);
209 atomic_set(&t->holders, 0);
210
211 if (!num_targets)
212 num_targets = KEYS_PER_NODE;
213
214 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
215
216 if (alloc_targets(t, num_targets)) {
217 kfree(t);
218 t = NULL;
219 return -ENOMEM;
220 }
221
222 t->mode = mode;
223 t->md = md;
224 *result = t;
225 return 0;
226}
227
228static void free_devices(struct list_head *devices)
229{
230 struct list_head *tmp, *next;
231
232 list_for_each_safe(tmp, next, devices) {
233 struct dm_dev_internal *dd =
234 list_entry(tmp, struct dm_dev_internal, list);
235 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
236 dd->dm_dev.name);
237 kfree(dd);
238 }
239}
240
241void dm_table_destroy(struct dm_table *t)
242{
243 unsigned int i;
244
245 if (!t)
246 return;
247
248 while (atomic_read(&t->holders))
249 msleep(1);
250 smp_mb();
251
252 /* free the indexes */
253 if (t->depth >= 2)
254 vfree(t->index[t->depth - 2]);
255
256 /* free the targets */
257 for (i = 0; i < t->num_targets; i++) {
258 struct dm_target *tgt = t->targets + i;
259
260 if (tgt->type->dtr)
261 tgt->type->dtr(tgt);
262
263 dm_put_target_type(tgt->type);
264 }
265
266 vfree(t->highs);
267
268 /* free the device list */
269 if (t->devices.next != &t->devices)
270 free_devices(&t->devices);
271
272 dm_free_md_mempools(t->mempools);
273
274 kfree(t);
275}
276
277void dm_table_get(struct dm_table *t)
278{
279 atomic_inc(&t->holders);
280}
281EXPORT_SYMBOL(dm_table_get);
282
283void dm_table_put(struct dm_table *t)
284{
285 if (!t)
286 return;
287
288 smp_mb__before_atomic_dec();
289 atomic_dec(&t->holders);
290}
291EXPORT_SYMBOL(dm_table_put);
292
293/*
294 * Checks to see if we need to extend highs or targets.
295 */
296static inline int check_space(struct dm_table *t)
297{
298 if (t->num_targets >= t->num_allocated)
299 return alloc_targets(t, t->num_allocated * 2);
300
301 return 0;
302}
303
304/*
305 * See if we've already got a device in the list.
306 */
307static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
308{
309 struct dm_dev_internal *dd;
310
311 list_for_each_entry (dd, l, list)
312 if (dd->dm_dev.bdev->bd_dev == dev)
313 return dd;
314
315 return NULL;
316}
317
318/*
319 * Open a device so we can use it as a map destination.
320 */
321static int open_dev(struct dm_dev_internal *d, dev_t dev,
322 struct mapped_device *md)
323{
324 static char *_claim_ptr = "I belong to device-mapper";
325 struct block_device *bdev;
326
327 int r;
328
329 BUG_ON(d->dm_dev.bdev);
330
331 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
332 if (IS_ERR(bdev))
333 return PTR_ERR(bdev);
334
335 r = bd_link_disk_holder(bdev, dm_disk(md));
336 if (r) {
337 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
338 return r;
339 }
340
341 d->dm_dev.bdev = bdev;
342 return 0;
343}
344
345/*
346 * Close a device that we've been using.
347 */
348static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
349{
350 if (!d->dm_dev.bdev)
351 return;
352
353 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
354 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
355 d->dm_dev.bdev = NULL;
356}
357
358/*
359 * If possible, this checks an area of a destination device is invalid.
360 */
361static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
362 sector_t start, sector_t len, void *data)
363{
364 struct request_queue *q;
365 struct queue_limits *limits = data;
366 struct block_device *bdev = dev->bdev;
367 sector_t dev_size =
368 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
369 unsigned short logical_block_size_sectors =
370 limits->logical_block_size >> SECTOR_SHIFT;
371 char b[BDEVNAME_SIZE];
372
373 /*
374 * Some devices exist without request functions,
375 * such as loop devices not yet bound to backing files.
376 * Forbid the use of such devices.
377 */
378 q = bdev_get_queue(bdev);
379 if (!q || !q->make_request_fn) {
380 DMWARN("%s: %s is not yet initialised: "
381 "start=%llu, len=%llu, dev_size=%llu",
382 dm_device_name(ti->table->md), bdevname(bdev, b),
383 (unsigned long long)start,
384 (unsigned long long)len,
385 (unsigned long long)dev_size);
386 return 1;
387 }
388
389 if (!dev_size)
390 return 0;
391
392 if ((start >= dev_size) || (start + len > dev_size)) {
393 DMWARN("%s: %s too small for target: "
394 "start=%llu, len=%llu, dev_size=%llu",
395 dm_device_name(ti->table->md), bdevname(bdev, b),
396 (unsigned long long)start,
397 (unsigned long long)len,
398 (unsigned long long)dev_size);
399 return 1;
400 }
401
402 if (logical_block_size_sectors <= 1)
403 return 0;
404
405 if (start & (logical_block_size_sectors - 1)) {
406 DMWARN("%s: start=%llu not aligned to h/w "
407 "logical block size %u of %s",
408 dm_device_name(ti->table->md),
409 (unsigned long long)start,
410 limits->logical_block_size, bdevname(bdev, b));
411 return 1;
412 }
413
414 if (len & (logical_block_size_sectors - 1)) {
415 DMWARN("%s: len=%llu not aligned to h/w "
416 "logical block size %u of %s",
417 dm_device_name(ti->table->md),
418 (unsigned long long)len,
419 limits->logical_block_size, bdevname(bdev, b));
420 return 1;
421 }
422
423 return 0;
424}
425
426/*
427 * This upgrades the mode on an already open dm_dev, being
428 * careful to leave things as they were if we fail to reopen the
429 * device and not to touch the existing bdev field in case
430 * it is accessed concurrently inside dm_table_any_congested().
431 */
432static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
433 struct mapped_device *md)
434{
435 int r;
436 struct dm_dev_internal dd_new, dd_old;
437
438 dd_new = dd_old = *dd;
439
440 dd_new.dm_dev.mode |= new_mode;
441 dd_new.dm_dev.bdev = NULL;
442
443 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
444 if (r)
445 return r;
446
447 dd->dm_dev.mode |= new_mode;
448 close_dev(&dd_old, md);
449
450 return 0;
451}
452
453/*
454 * Add a device to the list, or just increment the usage count if
455 * it's already present.
456 */
457int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
458 struct dm_dev **result)
459{
460 int r;
461 dev_t uninitialized_var(dev);
462 struct dm_dev_internal *dd;
463 unsigned int major, minor;
464 struct dm_table *t = ti->table;
465
466 BUG_ON(!t);
467
468 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
469 /* Extract the major/minor numbers */
470 dev = MKDEV(major, minor);
471 if (MAJOR(dev) != major || MINOR(dev) != minor)
472 return -EOVERFLOW;
473 } else {
474 /* convert the path to a device */
475 struct block_device *bdev = lookup_bdev(path);
476
477 if (IS_ERR(bdev))
478 return PTR_ERR(bdev);
479 dev = bdev->bd_dev;
480 bdput(bdev);
481 }
482
483 dd = find_device(&t->devices, dev);
484 if (!dd) {
485 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
486 if (!dd)
487 return -ENOMEM;
488
489 dd->dm_dev.mode = mode;
490 dd->dm_dev.bdev = NULL;
491
492 if ((r = open_dev(dd, dev, t->md))) {
493 kfree(dd);
494 return r;
495 }
496
497 format_dev_t(dd->dm_dev.name, dev);
498
499 atomic_set(&dd->count, 0);
500 list_add(&dd->list, &t->devices);
501
502 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
503 r = upgrade_mode(dd, mode, t->md);
504 if (r)
505 return r;
506 }
507 atomic_inc(&dd->count);
508
509 *result = &dd->dm_dev;
510 return 0;
511}
512EXPORT_SYMBOL(dm_get_device);
513
514int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
515 sector_t start, sector_t len, void *data)
516{
517 struct queue_limits *limits = data;
518 struct block_device *bdev = dev->bdev;
519 struct request_queue *q = bdev_get_queue(bdev);
520 char b[BDEVNAME_SIZE];
521
522 if (unlikely(!q)) {
523 DMWARN("%s: Cannot set limits for nonexistent device %s",
524 dm_device_name(ti->table->md), bdevname(bdev, b));
525 return 0;
526 }
527
528 if (bdev_stack_limits(limits, bdev, start) < 0)
529 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
530 "physical_block_size=%u, logical_block_size=%u, "
531 "alignment_offset=%u, start=%llu",
532 dm_device_name(ti->table->md), bdevname(bdev, b),
533 q->limits.physical_block_size,
534 q->limits.logical_block_size,
535 q->limits.alignment_offset,
536 (unsigned long long) start << SECTOR_SHIFT);
537
538 /*
539 * Check if merge fn is supported.
540 * If not we'll force DM to use PAGE_SIZE or
541 * smaller I/O, just to be safe.
542 */
543 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
544 blk_limits_max_hw_sectors(limits,
545 (unsigned int) (PAGE_SIZE >> 9));
546 return 0;
547}
548EXPORT_SYMBOL_GPL(dm_set_device_limits);
549
550/*
551 * Decrement a device's use count and remove it if necessary.
552 */
553void dm_put_device(struct dm_target *ti, struct dm_dev *d)
554{
555 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
556 dm_dev);
557
558 if (atomic_dec_and_test(&dd->count)) {
559 close_dev(dd, ti->table->md);
560 list_del(&dd->list);
561 kfree(dd);
562 }
563}
564EXPORT_SYMBOL(dm_put_device);
565
566/*
567 * Checks to see if the target joins onto the end of the table.
568 */
569static int adjoin(struct dm_table *table, struct dm_target *ti)
570{
571 struct dm_target *prev;
572
573 if (!table->num_targets)
574 return !ti->begin;
575
576 prev = &table->targets[table->num_targets - 1];
577 return (ti->begin == (prev->begin + prev->len));
578}
579
580/*
581 * Used to dynamically allocate the arg array.
582 */
583static char **realloc_argv(unsigned *array_size, char **old_argv)
584{
585 char **argv;
586 unsigned new_size;
587
588 new_size = *array_size ? *array_size * 2 : 64;
589 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
590 if (argv) {
591 memcpy(argv, old_argv, *array_size * sizeof(*argv));
592 *array_size = new_size;
593 }
594
595 kfree(old_argv);
596 return argv;
597}
598
599/*
600 * Destructively splits up the argument list to pass to ctr.
601 */
602int dm_split_args(int *argc, char ***argvp, char *input)
603{
604 char *start, *end = input, *out, **argv = NULL;
605 unsigned array_size = 0;
606
607 *argc = 0;
608
609 if (!input) {
610 *argvp = NULL;
611 return 0;
612 }
613
614 argv = realloc_argv(&array_size, argv);
615 if (!argv)
616 return -ENOMEM;
617
618 while (1) {
619 /* Skip whitespace */
620 start = skip_spaces(end);
621
622 if (!*start)
623 break; /* success, we hit the end */
624
625 /* 'out' is used to remove any back-quotes */
626 end = out = start;
627 while (*end) {
628 /* Everything apart from '\0' can be quoted */
629 if (*end == '\\' && *(end + 1)) {
630 *out++ = *(end + 1);
631 end += 2;
632 continue;
633 }
634
635 if (isspace(*end))
636 break; /* end of token */
637
638 *out++ = *end++;
639 }
640
641 /* have we already filled the array ? */
642 if ((*argc + 1) > array_size) {
643 argv = realloc_argv(&array_size, argv);
644 if (!argv)
645 return -ENOMEM;
646 }
647
648 /* we know this is whitespace */
649 if (*end)
650 end++;
651
652 /* terminate the string and put it in the array */
653 *out = '\0';
654 argv[*argc] = start;
655 (*argc)++;
656 }
657
658 *argvp = argv;
659 return 0;
660}
661
662/*
663 * Impose necessary and sufficient conditions on a devices's table such
664 * that any incoming bio which respects its logical_block_size can be
665 * processed successfully. If it falls across the boundary between
666 * two or more targets, the size of each piece it gets split into must
667 * be compatible with the logical_block_size of the target processing it.
668 */
669static int validate_hardware_logical_block_alignment(struct dm_table *table,
670 struct queue_limits *limits)
671{
672 /*
673 * This function uses arithmetic modulo the logical_block_size
674 * (in units of 512-byte sectors).
675 */
676 unsigned short device_logical_block_size_sects =
677 limits->logical_block_size >> SECTOR_SHIFT;
678
679 /*
680 * Offset of the start of the next table entry, mod logical_block_size.
681 */
682 unsigned short next_target_start = 0;
683
684 /*
685 * Given an aligned bio that extends beyond the end of a
686 * target, how many sectors must the next target handle?
687 */
688 unsigned short remaining = 0;
689
690 struct dm_target *uninitialized_var(ti);
691 struct queue_limits ti_limits;
692 unsigned i = 0;
693
694 /*
695 * Check each entry in the table in turn.
696 */
697 while (i < dm_table_get_num_targets(table)) {
698 ti = dm_table_get_target(table, i++);
699
700 blk_set_default_limits(&ti_limits);
701
702 /* combine all target devices' limits */
703 if (ti->type->iterate_devices)
704 ti->type->iterate_devices(ti, dm_set_device_limits,
705 &ti_limits);
706
707 /*
708 * If the remaining sectors fall entirely within this
709 * table entry are they compatible with its logical_block_size?
710 */
711 if (remaining < ti->len &&
712 remaining & ((ti_limits.logical_block_size >>
713 SECTOR_SHIFT) - 1))
714 break; /* Error */
715
716 next_target_start =
717 (unsigned short) ((next_target_start + ti->len) &
718 (device_logical_block_size_sects - 1));
719 remaining = next_target_start ?
720 device_logical_block_size_sects - next_target_start : 0;
721 }
722
723 if (remaining) {
724 DMWARN("%s: table line %u (start sect %llu len %llu) "
725 "not aligned to h/w logical block size %u",
726 dm_device_name(table->md), i,
727 (unsigned long long) ti->begin,
728 (unsigned long long) ti->len,
729 limits->logical_block_size);
730 return -EINVAL;
731 }
732
733 return 0;
734}
735
736int dm_table_add_target(struct dm_table *t, const char *type,
737 sector_t start, sector_t len, char *params)
738{
739 int r = -EINVAL, argc;
740 char **argv;
741 struct dm_target *tgt;
742
743 if ((r = check_space(t)))
744 return r;
745
746 tgt = t->targets + t->num_targets;
747 memset(tgt, 0, sizeof(*tgt));
748
749 if (!len) {
750 DMERR("%s: zero-length target", dm_device_name(t->md));
751 return -EINVAL;
752 }
753
754 tgt->type = dm_get_target_type(type);
755 if (!tgt->type) {
756 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
757 type);
758 return -EINVAL;
759 }
760
761 tgt->table = t;
762 tgt->begin = start;
763 tgt->len = len;
764 tgt->error = "Unknown error";
765
766 /*
767 * Does this target adjoin the previous one ?
768 */
769 if (!adjoin(t, tgt)) {
770 tgt->error = "Gap in table";
771 r = -EINVAL;
772 goto bad;
773 }
774
775 r = dm_split_args(&argc, &argv, params);
776 if (r) {
777 tgt->error = "couldn't split parameters (insufficient memory)";
778 goto bad;
779 }
780
781 r = tgt->type->ctr(tgt, argc, argv);
782 kfree(argv);
783 if (r)
784 goto bad;
785
786 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
787
788 if (!tgt->num_discard_requests && tgt->discards_supported)
789 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
790 dm_device_name(t->md), type);
791
792 return 0;
793
794 bad:
795 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
796 dm_put_target_type(tgt->type);
797 return r;
798}
799
800/*
801 * Target argument parsing helpers.
802 */
803static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
804 unsigned *value, char **error, unsigned grouped)
805{
806 const char *arg_str = dm_shift_arg(arg_set);
807
808 if (!arg_str ||
809 (sscanf(arg_str, "%u", value) != 1) ||
810 (*value < arg->min) ||
811 (*value > arg->max) ||
812 (grouped && arg_set->argc < *value)) {
813 *error = arg->error;
814 return -EINVAL;
815 }
816
817 return 0;
818}
819
820int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
821 unsigned *value, char **error)
822{
823 return validate_next_arg(arg, arg_set, value, error, 0);
824}
825EXPORT_SYMBOL(dm_read_arg);
826
827int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
828 unsigned *value, char **error)
829{
830 return validate_next_arg(arg, arg_set, value, error, 1);
831}
832EXPORT_SYMBOL(dm_read_arg_group);
833
834const char *dm_shift_arg(struct dm_arg_set *as)
835{
836 char *r;
837
838 if (as->argc) {
839 as->argc--;
840 r = *as->argv;
841 as->argv++;
842 return r;
843 }
844
845 return NULL;
846}
847EXPORT_SYMBOL(dm_shift_arg);
848
849void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
850{
851 BUG_ON(as->argc < num_args);
852 as->argc -= num_args;
853 as->argv += num_args;
854}
855EXPORT_SYMBOL(dm_consume_args);
856
857static int dm_table_set_type(struct dm_table *t)
858{
859 unsigned i;
860 unsigned bio_based = 0, request_based = 0;
861 struct dm_target *tgt;
862 struct dm_dev_internal *dd;
863 struct list_head *devices;
864
865 for (i = 0; i < t->num_targets; i++) {
866 tgt = t->targets + i;
867 if (dm_target_request_based(tgt))
868 request_based = 1;
869 else
870 bio_based = 1;
871
872 if (bio_based && request_based) {
873 DMWARN("Inconsistent table: different target types"
874 " can't be mixed up");
875 return -EINVAL;
876 }
877 }
878
879 if (bio_based) {
880 /* We must use this table as bio-based */
881 t->type = DM_TYPE_BIO_BASED;
882 return 0;
883 }
884
885 BUG_ON(!request_based); /* No targets in this table */
886
887 /* Non-request-stackable devices can't be used for request-based dm */
888 devices = dm_table_get_devices(t);
889 list_for_each_entry(dd, devices, list) {
890 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
891 DMWARN("table load rejected: including"
892 " non-request-stackable devices");
893 return -EINVAL;
894 }
895 }
896
897 /*
898 * Request-based dm supports only tables that have a single target now.
899 * To support multiple targets, request splitting support is needed,
900 * and that needs lots of changes in the block-layer.
901 * (e.g. request completion process for partial completion.)
902 */
903 if (t->num_targets > 1) {
904 DMWARN("Request-based dm doesn't support multiple targets yet");
905 return -EINVAL;
906 }
907
908 t->type = DM_TYPE_REQUEST_BASED;
909
910 return 0;
911}
912
913unsigned dm_table_get_type(struct dm_table *t)
914{
915 return t->type;
916}
917
918bool dm_table_request_based(struct dm_table *t)
919{
920 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
921}
922
923int dm_table_alloc_md_mempools(struct dm_table *t)
924{
925 unsigned type = dm_table_get_type(t);
926
927 if (unlikely(type == DM_TYPE_NONE)) {
928 DMWARN("no table type is set, can't allocate mempools");
929 return -EINVAL;
930 }
931
932 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
933 if (!t->mempools)
934 return -ENOMEM;
935
936 return 0;
937}
938
939void dm_table_free_md_mempools(struct dm_table *t)
940{
941 dm_free_md_mempools(t->mempools);
942 t->mempools = NULL;
943}
944
945struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
946{
947 return t->mempools;
948}
949
950static int setup_indexes(struct dm_table *t)
951{
952 int i;
953 unsigned int total = 0;
954 sector_t *indexes;
955
956 /* allocate the space for *all* the indexes */
957 for (i = t->depth - 2; i >= 0; i--) {
958 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
959 total += t->counts[i];
960 }
961
962 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
963 if (!indexes)
964 return -ENOMEM;
965
966 /* set up internal nodes, bottom-up */
967 for (i = t->depth - 2; i >= 0; i--) {
968 t->index[i] = indexes;
969 indexes += (KEYS_PER_NODE * t->counts[i]);
970 setup_btree_index(i, t);
971 }
972
973 return 0;
974}
975
976/*
977 * Builds the btree to index the map.
978 */
979static int dm_table_build_index(struct dm_table *t)
980{
981 int r = 0;
982 unsigned int leaf_nodes;
983
984 /* how many indexes will the btree have ? */
985 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
986 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
987
988 /* leaf layer has already been set up */
989 t->counts[t->depth - 1] = leaf_nodes;
990 t->index[t->depth - 1] = t->highs;
991
992 if (t->depth >= 2)
993 r = setup_indexes(t);
994
995 return r;
996}
997
998/*
999 * Get a disk whose integrity profile reflects the table's profile.
1000 * If %match_all is true, all devices' profiles must match.
1001 * If %match_all is false, all devices must at least have an
1002 * allocated integrity profile; but uninitialized is ok.
1003 * Returns NULL if integrity support was inconsistent or unavailable.
1004 */
1005static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1006 bool match_all)
1007{
1008 struct list_head *devices = dm_table_get_devices(t);
1009 struct dm_dev_internal *dd = NULL;
1010 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1011
1012 list_for_each_entry(dd, devices, list) {
1013 template_disk = dd->dm_dev.bdev->bd_disk;
1014 if (!blk_get_integrity(template_disk))
1015 goto no_integrity;
1016 if (!match_all && !blk_integrity_is_initialized(template_disk))
1017 continue; /* skip uninitialized profiles */
1018 else if (prev_disk &&
1019 blk_integrity_compare(prev_disk, template_disk) < 0)
1020 goto no_integrity;
1021 prev_disk = template_disk;
1022 }
1023
1024 return template_disk;
1025
1026no_integrity:
1027 if (prev_disk)
1028 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1029 dm_device_name(t->md),
1030 prev_disk->disk_name,
1031 template_disk->disk_name);
1032 return NULL;
1033}
1034
1035/*
1036 * Register the mapped device for blk_integrity support if
1037 * the underlying devices have an integrity profile. But all devices
1038 * may not have matching profiles (checking all devices isn't reliable
1039 * during table load because this table may use other DM device(s) which
1040 * must be resumed before they will have an initialized integity profile).
1041 * Stacked DM devices force a 2 stage integrity profile validation:
1042 * 1 - during load, validate all initialized integrity profiles match
1043 * 2 - during resume, validate all integrity profiles match
1044 */
1045static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1046{
1047 struct gendisk *template_disk = NULL;
1048
1049 template_disk = dm_table_get_integrity_disk(t, false);
1050 if (!template_disk)
1051 return 0;
1052
1053 if (!blk_integrity_is_initialized(dm_disk(md))) {
1054 t->integrity_supported = 1;
1055 return blk_integrity_register(dm_disk(md), NULL);
1056 }
1057
1058 /*
1059 * If DM device already has an initalized integrity
1060 * profile the new profile should not conflict.
1061 */
1062 if (blk_integrity_is_initialized(template_disk) &&
1063 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1064 DMWARN("%s: conflict with existing integrity profile: "
1065 "%s profile mismatch",
1066 dm_device_name(t->md),
1067 template_disk->disk_name);
1068 return 1;
1069 }
1070
1071 /* Preserve existing initialized integrity profile */
1072 t->integrity_supported = 1;
1073 return 0;
1074}
1075
1076/*
1077 * Prepares the table for use by building the indices,
1078 * setting the type, and allocating mempools.
1079 */
1080int dm_table_complete(struct dm_table *t)
1081{
1082 int r;
1083
1084 r = dm_table_set_type(t);
1085 if (r) {
1086 DMERR("unable to set table type");
1087 return r;
1088 }
1089
1090 r = dm_table_build_index(t);
1091 if (r) {
1092 DMERR("unable to build btrees");
1093 return r;
1094 }
1095
1096 r = dm_table_prealloc_integrity(t, t->md);
1097 if (r) {
1098 DMERR("could not register integrity profile.");
1099 return r;
1100 }
1101
1102 r = dm_table_alloc_md_mempools(t);
1103 if (r)
1104 DMERR("unable to allocate mempools");
1105
1106 return r;
1107}
1108
1109static DEFINE_MUTEX(_event_lock);
1110void dm_table_event_callback(struct dm_table *t,
1111 void (*fn)(void *), void *context)
1112{
1113 mutex_lock(&_event_lock);
1114 t->event_fn = fn;
1115 t->event_context = context;
1116 mutex_unlock(&_event_lock);
1117}
1118
1119void dm_table_event(struct dm_table *t)
1120{
1121 /*
1122 * You can no longer call dm_table_event() from interrupt
1123 * context, use a bottom half instead.
1124 */
1125 BUG_ON(in_interrupt());
1126
1127 mutex_lock(&_event_lock);
1128 if (t->event_fn)
1129 t->event_fn(t->event_context);
1130 mutex_unlock(&_event_lock);
1131}
1132EXPORT_SYMBOL(dm_table_event);
1133
1134sector_t dm_table_get_size(struct dm_table *t)
1135{
1136 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1137}
1138EXPORT_SYMBOL(dm_table_get_size);
1139
1140struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1141{
1142 if (index >= t->num_targets)
1143 return NULL;
1144
1145 return t->targets + index;
1146}
1147
1148/*
1149 * Search the btree for the correct target.
1150 *
1151 * Caller should check returned pointer with dm_target_is_valid()
1152 * to trap I/O beyond end of device.
1153 */
1154struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1155{
1156 unsigned int l, n = 0, k = 0;
1157 sector_t *node;
1158
1159 for (l = 0; l < t->depth; l++) {
1160 n = get_child(n, k);
1161 node = get_node(t, l, n);
1162
1163 for (k = 0; k < KEYS_PER_NODE; k++)
1164 if (node[k] >= sector)
1165 break;
1166 }
1167
1168 return &t->targets[(KEYS_PER_NODE * n) + k];
1169}
1170
1171/*
1172 * Establish the new table's queue_limits and validate them.
1173 */
1174int dm_calculate_queue_limits(struct dm_table *table,
1175 struct queue_limits *limits)
1176{
1177 struct dm_target *uninitialized_var(ti);
1178 struct queue_limits ti_limits;
1179 unsigned i = 0;
1180
1181 blk_set_default_limits(limits);
1182
1183 while (i < dm_table_get_num_targets(table)) {
1184 blk_set_default_limits(&ti_limits);
1185
1186 ti = dm_table_get_target(table, i++);
1187
1188 if (!ti->type->iterate_devices)
1189 goto combine_limits;
1190
1191 /*
1192 * Combine queue limits of all the devices this target uses.
1193 */
1194 ti->type->iterate_devices(ti, dm_set_device_limits,
1195 &ti_limits);
1196
1197 /* Set I/O hints portion of queue limits */
1198 if (ti->type->io_hints)
1199 ti->type->io_hints(ti, &ti_limits);
1200
1201 /*
1202 * Check each device area is consistent with the target's
1203 * overall queue limits.
1204 */
1205 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1206 &ti_limits))
1207 return -EINVAL;
1208
1209combine_limits:
1210 /*
1211 * Merge this target's queue limits into the overall limits
1212 * for the table.
1213 */
1214 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1215 DMWARN("%s: adding target device "
1216 "(start sect %llu len %llu) "
1217 "caused an alignment inconsistency",
1218 dm_device_name(table->md),
1219 (unsigned long long) ti->begin,
1220 (unsigned long long) ti->len);
1221 }
1222
1223 return validate_hardware_logical_block_alignment(table, limits);
1224}
1225
1226/*
1227 * Set the integrity profile for this device if all devices used have
1228 * matching profiles. We're quite deep in the resume path but still
1229 * don't know if all devices (particularly DM devices this device
1230 * may be stacked on) have matching profiles. Even if the profiles
1231 * don't match we have no way to fail (to resume) at this point.
1232 */
1233static void dm_table_set_integrity(struct dm_table *t)
1234{
1235 struct gendisk *template_disk = NULL;
1236
1237 if (!blk_get_integrity(dm_disk(t->md)))
1238 return;
1239
1240 template_disk = dm_table_get_integrity_disk(t, true);
1241 if (template_disk)
1242 blk_integrity_register(dm_disk(t->md),
1243 blk_get_integrity(template_disk));
1244 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1245 DMWARN("%s: device no longer has a valid integrity profile",
1246 dm_device_name(t->md));
1247 else
1248 DMWARN("%s: unable to establish an integrity profile",
1249 dm_device_name(t->md));
1250}
1251
1252static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1253 sector_t start, sector_t len, void *data)
1254{
1255 unsigned flush = (*(unsigned *)data);
1256 struct request_queue *q = bdev_get_queue(dev->bdev);
1257
1258 return q && (q->flush_flags & flush);
1259}
1260
1261static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1262{
1263 struct dm_target *ti;
1264 unsigned i = 0;
1265
1266 /*
1267 * Require at least one underlying device to support flushes.
1268 * t->devices includes internal dm devices such as mirror logs
1269 * so we need to use iterate_devices here, which targets
1270 * supporting flushes must provide.
1271 */
1272 while (i < dm_table_get_num_targets(t)) {
1273 ti = dm_table_get_target(t, i++);
1274
1275 if (!ti->num_flush_requests)
1276 continue;
1277
1278 if (ti->type->iterate_devices &&
1279 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1280 return 1;
1281 }
1282
1283 return 0;
1284}
1285
1286static bool dm_table_discard_zeroes_data(struct dm_table *t)
1287{
1288 struct dm_target *ti;
1289 unsigned i = 0;
1290
1291 /* Ensure that all targets supports discard_zeroes_data. */
1292 while (i < dm_table_get_num_targets(t)) {
1293 ti = dm_table_get_target(t, i++);
1294
1295 if (ti->discard_zeroes_data_unsupported)
1296 return 0;
1297 }
1298
1299 return 1;
1300}
1301
1302void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1303 struct queue_limits *limits)
1304{
1305 unsigned flush = 0;
1306
1307 /*
1308 * Copy table's limits to the DM device's request_queue
1309 */
1310 q->limits = *limits;
1311
1312 if (!dm_table_supports_discards(t))
1313 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1314 else
1315 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1316
1317 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1318 flush |= REQ_FLUSH;
1319 if (dm_table_supports_flush(t, REQ_FUA))
1320 flush |= REQ_FUA;
1321 }
1322 blk_queue_flush(q, flush);
1323
1324 if (!dm_table_discard_zeroes_data(t))
1325 q->limits.discard_zeroes_data = 0;
1326
1327 dm_table_set_integrity(t);
1328
1329 /*
1330 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1331 * visible to other CPUs because, once the flag is set, incoming bios
1332 * are processed by request-based dm, which refers to the queue
1333 * settings.
1334 * Until the flag set, bios are passed to bio-based dm and queued to
1335 * md->deferred where queue settings are not needed yet.
1336 * Those bios are passed to request-based dm at the resume time.
1337 */
1338 smp_mb();
1339 if (dm_table_request_based(t))
1340 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1341}
1342
1343unsigned int dm_table_get_num_targets(struct dm_table *t)
1344{
1345 return t->num_targets;
1346}
1347
1348struct list_head *dm_table_get_devices(struct dm_table *t)
1349{
1350 return &t->devices;
1351}
1352
1353fmode_t dm_table_get_mode(struct dm_table *t)
1354{
1355 return t->mode;
1356}
1357EXPORT_SYMBOL(dm_table_get_mode);
1358
1359static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1360{
1361 int i = t->num_targets;
1362 struct dm_target *ti = t->targets;
1363
1364 while (i--) {
1365 if (postsuspend) {
1366 if (ti->type->postsuspend)
1367 ti->type->postsuspend(ti);
1368 } else if (ti->type->presuspend)
1369 ti->type->presuspend(ti);
1370
1371 ti++;
1372 }
1373}
1374
1375void dm_table_presuspend_targets(struct dm_table *t)
1376{
1377 if (!t)
1378 return;
1379
1380 suspend_targets(t, 0);
1381}
1382
1383void dm_table_postsuspend_targets(struct dm_table *t)
1384{
1385 if (!t)
1386 return;
1387
1388 suspend_targets(t, 1);
1389}
1390
1391int dm_table_resume_targets(struct dm_table *t)
1392{
1393 int i, r = 0;
1394
1395 for (i = 0; i < t->num_targets; i++) {
1396 struct dm_target *ti = t->targets + i;
1397
1398 if (!ti->type->preresume)
1399 continue;
1400
1401 r = ti->type->preresume(ti);
1402 if (r)
1403 return r;
1404 }
1405
1406 for (i = 0; i < t->num_targets; i++) {
1407 struct dm_target *ti = t->targets + i;
1408
1409 if (ti->type->resume)
1410 ti->type->resume(ti);
1411 }
1412
1413 return 0;
1414}
1415
1416void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1417{
1418 list_add(&cb->list, &t->target_callbacks);
1419}
1420EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1421
1422int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1423{
1424 struct dm_dev_internal *dd;
1425 struct list_head *devices = dm_table_get_devices(t);
1426 struct dm_target_callbacks *cb;
1427 int r = 0;
1428
1429 list_for_each_entry(dd, devices, list) {
1430 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1431 char b[BDEVNAME_SIZE];
1432
1433 if (likely(q))
1434 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1435 else
1436 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1437 dm_device_name(t->md),
1438 bdevname(dd->dm_dev.bdev, b));
1439 }
1440
1441 list_for_each_entry(cb, &t->target_callbacks, list)
1442 if (cb->congested_fn)
1443 r |= cb->congested_fn(cb, bdi_bits);
1444
1445 return r;
1446}
1447
1448int dm_table_any_busy_target(struct dm_table *t)
1449{
1450 unsigned i;
1451 struct dm_target *ti;
1452
1453 for (i = 0; i < t->num_targets; i++) {
1454 ti = t->targets + i;
1455 if (ti->type->busy && ti->type->busy(ti))
1456 return 1;
1457 }
1458
1459 return 0;
1460}
1461
1462struct mapped_device *dm_table_get_md(struct dm_table *t)
1463{
1464 return t->md;
1465}
1466EXPORT_SYMBOL(dm_table_get_md);
1467
1468static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1469 sector_t start, sector_t len, void *data)
1470{
1471 struct request_queue *q = bdev_get_queue(dev->bdev);
1472
1473 return q && blk_queue_discard(q);
1474}
1475
1476bool dm_table_supports_discards(struct dm_table *t)
1477{
1478 struct dm_target *ti;
1479 unsigned i = 0;
1480
1481 /*
1482 * Unless any target used by the table set discards_supported,
1483 * require at least one underlying device to support discards.
1484 * t->devices includes internal dm devices such as mirror logs
1485 * so we need to use iterate_devices here, which targets
1486 * supporting discard selectively must provide.
1487 */
1488 while (i < dm_table_get_num_targets(t)) {
1489 ti = dm_table_get_target(t, i++);
1490
1491 if (!ti->num_discard_requests)
1492 continue;
1493
1494 if (ti->discards_supported)
1495 return 1;
1496
1497 if (ti->type->iterate_devices &&
1498 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1499 return 1;
1500 }
1501
1502 return 0;
1503}
1/*
2 * Copyright (C) 2001 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include "dm-core.h"
9
10#include <linux/module.h>
11#include <linux/vmalloc.h>
12#include <linux/blkdev.h>
13#include <linux/namei.h>
14#include <linux/ctype.h>
15#include <linux/string.h>
16#include <linux/slab.h>
17#include <linux/interrupt.h>
18#include <linux/mutex.h>
19#include <linux/delay.h>
20#include <linux/atomic.h>
21#include <linux/blk-mq.h>
22#include <linux/mount.h>
23
24#define DM_MSG_PREFIX "table"
25
26#define MAX_DEPTH 16
27#define NODE_SIZE L1_CACHE_BYTES
28#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
29#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
30
31struct dm_table {
32 struct mapped_device *md;
33 unsigned type;
34
35 /* btree table */
36 unsigned int depth;
37 unsigned int counts[MAX_DEPTH]; /* in nodes */
38 sector_t *index[MAX_DEPTH];
39
40 unsigned int num_targets;
41 unsigned int num_allocated;
42 sector_t *highs;
43 struct dm_target *targets;
44
45 struct target_type *immutable_target_type;
46
47 bool integrity_supported:1;
48 bool singleton:1;
49 bool all_blk_mq:1;
50
51 /*
52 * Indicates the rw permissions for the new logical
53 * device. This should be a combination of FMODE_READ
54 * and FMODE_WRITE.
55 */
56 fmode_t mode;
57
58 /* a list of devices used by this table */
59 struct list_head devices;
60
61 /* events get handed up using this callback */
62 void (*event_fn)(void *);
63 void *event_context;
64
65 struct dm_md_mempools *mempools;
66
67 struct list_head target_callbacks;
68};
69
70/*
71 * Similar to ceiling(log_size(n))
72 */
73static unsigned int int_log(unsigned int n, unsigned int base)
74{
75 int result = 0;
76
77 while (n > 1) {
78 n = dm_div_up(n, base);
79 result++;
80 }
81
82 return result;
83}
84
85/*
86 * Calculate the index of the child node of the n'th node k'th key.
87 */
88static inline unsigned int get_child(unsigned int n, unsigned int k)
89{
90 return (n * CHILDREN_PER_NODE) + k;
91}
92
93/*
94 * Return the n'th node of level l from table t.
95 */
96static inline sector_t *get_node(struct dm_table *t,
97 unsigned int l, unsigned int n)
98{
99 return t->index[l] + (n * KEYS_PER_NODE);
100}
101
102/*
103 * Return the highest key that you could lookup from the n'th
104 * node on level l of the btree.
105 */
106static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
107{
108 for (; l < t->depth - 1; l++)
109 n = get_child(n, CHILDREN_PER_NODE - 1);
110
111 if (n >= t->counts[l])
112 return (sector_t) - 1;
113
114 return get_node(t, l, n)[KEYS_PER_NODE - 1];
115}
116
117/*
118 * Fills in a level of the btree based on the highs of the level
119 * below it.
120 */
121static int setup_btree_index(unsigned int l, struct dm_table *t)
122{
123 unsigned int n, k;
124 sector_t *node;
125
126 for (n = 0U; n < t->counts[l]; n++) {
127 node = get_node(t, l, n);
128
129 for (k = 0U; k < KEYS_PER_NODE; k++)
130 node[k] = high(t, l + 1, get_child(n, k));
131 }
132
133 return 0;
134}
135
136void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
137{
138 unsigned long size;
139 void *addr;
140
141 /*
142 * Check that we're not going to overflow.
143 */
144 if (nmemb > (ULONG_MAX / elem_size))
145 return NULL;
146
147 size = nmemb * elem_size;
148 addr = vzalloc(size);
149
150 return addr;
151}
152EXPORT_SYMBOL(dm_vcalloc);
153
154/*
155 * highs, and targets are managed as dynamic arrays during a
156 * table load.
157 */
158static int alloc_targets(struct dm_table *t, unsigned int num)
159{
160 sector_t *n_highs;
161 struct dm_target *n_targets;
162
163 /*
164 * Allocate both the target array and offset array at once.
165 * Append an empty entry to catch sectors beyond the end of
166 * the device.
167 */
168 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
169 sizeof(sector_t));
170 if (!n_highs)
171 return -ENOMEM;
172
173 n_targets = (struct dm_target *) (n_highs + num);
174
175 memset(n_highs, -1, sizeof(*n_highs) * num);
176 vfree(t->highs);
177
178 t->num_allocated = num;
179 t->highs = n_highs;
180 t->targets = n_targets;
181
182 return 0;
183}
184
185int dm_table_create(struct dm_table **result, fmode_t mode,
186 unsigned num_targets, struct mapped_device *md)
187{
188 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
189
190 if (!t)
191 return -ENOMEM;
192
193 INIT_LIST_HEAD(&t->devices);
194 INIT_LIST_HEAD(&t->target_callbacks);
195
196 if (!num_targets)
197 num_targets = KEYS_PER_NODE;
198
199 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
200
201 if (!num_targets) {
202 kfree(t);
203 return -ENOMEM;
204 }
205
206 if (alloc_targets(t, num_targets)) {
207 kfree(t);
208 return -ENOMEM;
209 }
210
211 t->type = DM_TYPE_NONE;
212 t->mode = mode;
213 t->md = md;
214 *result = t;
215 return 0;
216}
217
218static void free_devices(struct list_head *devices, struct mapped_device *md)
219{
220 struct list_head *tmp, *next;
221
222 list_for_each_safe(tmp, next, devices) {
223 struct dm_dev_internal *dd =
224 list_entry(tmp, struct dm_dev_internal, list);
225 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
226 dm_device_name(md), dd->dm_dev->name);
227 dm_put_table_device(md, dd->dm_dev);
228 kfree(dd);
229 }
230}
231
232void dm_table_destroy(struct dm_table *t)
233{
234 unsigned int i;
235
236 if (!t)
237 return;
238
239 /* free the indexes */
240 if (t->depth >= 2)
241 vfree(t->index[t->depth - 2]);
242
243 /* free the targets */
244 for (i = 0; i < t->num_targets; i++) {
245 struct dm_target *tgt = t->targets + i;
246
247 if (tgt->type->dtr)
248 tgt->type->dtr(tgt);
249
250 dm_put_target_type(tgt->type);
251 }
252
253 vfree(t->highs);
254
255 /* free the device list */
256 free_devices(&t->devices, t->md);
257
258 dm_free_md_mempools(t->mempools);
259
260 kfree(t);
261}
262
263/*
264 * See if we've already got a device in the list.
265 */
266static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
267{
268 struct dm_dev_internal *dd;
269
270 list_for_each_entry (dd, l, list)
271 if (dd->dm_dev->bdev->bd_dev == dev)
272 return dd;
273
274 return NULL;
275}
276
277/*
278 * If possible, this checks an area of a destination device is invalid.
279 */
280static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
281 sector_t start, sector_t len, void *data)
282{
283 struct request_queue *q;
284 struct queue_limits *limits = data;
285 struct block_device *bdev = dev->bdev;
286 sector_t dev_size =
287 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
288 unsigned short logical_block_size_sectors =
289 limits->logical_block_size >> SECTOR_SHIFT;
290 char b[BDEVNAME_SIZE];
291
292 /*
293 * Some devices exist without request functions,
294 * such as loop devices not yet bound to backing files.
295 * Forbid the use of such devices.
296 */
297 q = bdev_get_queue(bdev);
298 if (!q || !q->make_request_fn) {
299 DMWARN("%s: %s is not yet initialised: "
300 "start=%llu, len=%llu, dev_size=%llu",
301 dm_device_name(ti->table->md), bdevname(bdev, b),
302 (unsigned long long)start,
303 (unsigned long long)len,
304 (unsigned long long)dev_size);
305 return 1;
306 }
307
308 if (!dev_size)
309 return 0;
310
311 if ((start >= dev_size) || (start + len > dev_size)) {
312 DMWARN("%s: %s too small for target: "
313 "start=%llu, len=%llu, dev_size=%llu",
314 dm_device_name(ti->table->md), bdevname(bdev, b),
315 (unsigned long long)start,
316 (unsigned long long)len,
317 (unsigned long long)dev_size);
318 return 1;
319 }
320
321 if (logical_block_size_sectors <= 1)
322 return 0;
323
324 if (start & (logical_block_size_sectors - 1)) {
325 DMWARN("%s: start=%llu not aligned to h/w "
326 "logical block size %u of %s",
327 dm_device_name(ti->table->md),
328 (unsigned long long)start,
329 limits->logical_block_size, bdevname(bdev, b));
330 return 1;
331 }
332
333 if (len & (logical_block_size_sectors - 1)) {
334 DMWARN("%s: len=%llu not aligned to h/w "
335 "logical block size %u of %s",
336 dm_device_name(ti->table->md),
337 (unsigned long long)len,
338 limits->logical_block_size, bdevname(bdev, b));
339 return 1;
340 }
341
342 return 0;
343}
344
345/*
346 * This upgrades the mode on an already open dm_dev, being
347 * careful to leave things as they were if we fail to reopen the
348 * device and not to touch the existing bdev field in case
349 * it is accessed concurrently inside dm_table_any_congested().
350 */
351static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
352 struct mapped_device *md)
353{
354 int r;
355 struct dm_dev *old_dev, *new_dev;
356
357 old_dev = dd->dm_dev;
358
359 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
360 dd->dm_dev->mode | new_mode, &new_dev);
361 if (r)
362 return r;
363
364 dd->dm_dev = new_dev;
365 dm_put_table_device(md, old_dev);
366
367 return 0;
368}
369
370/*
371 * Convert the path to a device
372 */
373dev_t dm_get_dev_t(const char *path)
374{
375 dev_t uninitialized_var(dev);
376 struct block_device *bdev;
377
378 bdev = lookup_bdev(path);
379 if (IS_ERR(bdev))
380 dev = name_to_dev_t(path);
381 else {
382 dev = bdev->bd_dev;
383 bdput(bdev);
384 }
385
386 return dev;
387}
388EXPORT_SYMBOL_GPL(dm_get_dev_t);
389
390/*
391 * Add a device to the list, or just increment the usage count if
392 * it's already present.
393 */
394int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
395 struct dm_dev **result)
396{
397 int r;
398 dev_t dev;
399 struct dm_dev_internal *dd;
400 struct dm_table *t = ti->table;
401
402 BUG_ON(!t);
403
404 dev = dm_get_dev_t(path);
405 if (!dev)
406 return -ENODEV;
407
408 dd = find_device(&t->devices, dev);
409 if (!dd) {
410 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
411 if (!dd)
412 return -ENOMEM;
413
414 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
415 kfree(dd);
416 return r;
417 }
418
419 atomic_set(&dd->count, 0);
420 list_add(&dd->list, &t->devices);
421
422 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
423 r = upgrade_mode(dd, mode, t->md);
424 if (r)
425 return r;
426 }
427 atomic_inc(&dd->count);
428
429 *result = dd->dm_dev;
430 return 0;
431}
432EXPORT_SYMBOL(dm_get_device);
433
434static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
435 sector_t start, sector_t len, void *data)
436{
437 struct queue_limits *limits = data;
438 struct block_device *bdev = dev->bdev;
439 struct request_queue *q = bdev_get_queue(bdev);
440 char b[BDEVNAME_SIZE];
441
442 if (unlikely(!q)) {
443 DMWARN("%s: Cannot set limits for nonexistent device %s",
444 dm_device_name(ti->table->md), bdevname(bdev, b));
445 return 0;
446 }
447
448 if (bdev_stack_limits(limits, bdev, start) < 0)
449 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
450 "physical_block_size=%u, logical_block_size=%u, "
451 "alignment_offset=%u, start=%llu",
452 dm_device_name(ti->table->md), bdevname(bdev, b),
453 q->limits.physical_block_size,
454 q->limits.logical_block_size,
455 q->limits.alignment_offset,
456 (unsigned long long) start << SECTOR_SHIFT);
457
458 return 0;
459}
460
461/*
462 * Decrement a device's use count and remove it if necessary.
463 */
464void dm_put_device(struct dm_target *ti, struct dm_dev *d)
465{
466 int found = 0;
467 struct list_head *devices = &ti->table->devices;
468 struct dm_dev_internal *dd;
469
470 list_for_each_entry(dd, devices, list) {
471 if (dd->dm_dev == d) {
472 found = 1;
473 break;
474 }
475 }
476 if (!found) {
477 DMWARN("%s: device %s not in table devices list",
478 dm_device_name(ti->table->md), d->name);
479 return;
480 }
481 if (atomic_dec_and_test(&dd->count)) {
482 dm_put_table_device(ti->table->md, d);
483 list_del(&dd->list);
484 kfree(dd);
485 }
486}
487EXPORT_SYMBOL(dm_put_device);
488
489/*
490 * Checks to see if the target joins onto the end of the table.
491 */
492static int adjoin(struct dm_table *table, struct dm_target *ti)
493{
494 struct dm_target *prev;
495
496 if (!table->num_targets)
497 return !ti->begin;
498
499 prev = &table->targets[table->num_targets - 1];
500 return (ti->begin == (prev->begin + prev->len));
501}
502
503/*
504 * Used to dynamically allocate the arg array.
505 *
506 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
507 * process messages even if some device is suspended. These messages have a
508 * small fixed number of arguments.
509 *
510 * On the other hand, dm-switch needs to process bulk data using messages and
511 * excessive use of GFP_NOIO could cause trouble.
512 */
513static char **realloc_argv(unsigned *array_size, char **old_argv)
514{
515 char **argv;
516 unsigned new_size;
517 gfp_t gfp;
518
519 if (*array_size) {
520 new_size = *array_size * 2;
521 gfp = GFP_KERNEL;
522 } else {
523 new_size = 8;
524 gfp = GFP_NOIO;
525 }
526 argv = kmalloc(new_size * sizeof(*argv), gfp);
527 if (argv) {
528 memcpy(argv, old_argv, *array_size * sizeof(*argv));
529 *array_size = new_size;
530 }
531
532 kfree(old_argv);
533 return argv;
534}
535
536/*
537 * Destructively splits up the argument list to pass to ctr.
538 */
539int dm_split_args(int *argc, char ***argvp, char *input)
540{
541 char *start, *end = input, *out, **argv = NULL;
542 unsigned array_size = 0;
543
544 *argc = 0;
545
546 if (!input) {
547 *argvp = NULL;
548 return 0;
549 }
550
551 argv = realloc_argv(&array_size, argv);
552 if (!argv)
553 return -ENOMEM;
554
555 while (1) {
556 /* Skip whitespace */
557 start = skip_spaces(end);
558
559 if (!*start)
560 break; /* success, we hit the end */
561
562 /* 'out' is used to remove any back-quotes */
563 end = out = start;
564 while (*end) {
565 /* Everything apart from '\0' can be quoted */
566 if (*end == '\\' && *(end + 1)) {
567 *out++ = *(end + 1);
568 end += 2;
569 continue;
570 }
571
572 if (isspace(*end))
573 break; /* end of token */
574
575 *out++ = *end++;
576 }
577
578 /* have we already filled the array ? */
579 if ((*argc + 1) > array_size) {
580 argv = realloc_argv(&array_size, argv);
581 if (!argv)
582 return -ENOMEM;
583 }
584
585 /* we know this is whitespace */
586 if (*end)
587 end++;
588
589 /* terminate the string and put it in the array */
590 *out = '\0';
591 argv[*argc] = start;
592 (*argc)++;
593 }
594
595 *argvp = argv;
596 return 0;
597}
598
599/*
600 * Impose necessary and sufficient conditions on a devices's table such
601 * that any incoming bio which respects its logical_block_size can be
602 * processed successfully. If it falls across the boundary between
603 * two or more targets, the size of each piece it gets split into must
604 * be compatible with the logical_block_size of the target processing it.
605 */
606static int validate_hardware_logical_block_alignment(struct dm_table *table,
607 struct queue_limits *limits)
608{
609 /*
610 * This function uses arithmetic modulo the logical_block_size
611 * (in units of 512-byte sectors).
612 */
613 unsigned short device_logical_block_size_sects =
614 limits->logical_block_size >> SECTOR_SHIFT;
615
616 /*
617 * Offset of the start of the next table entry, mod logical_block_size.
618 */
619 unsigned short next_target_start = 0;
620
621 /*
622 * Given an aligned bio that extends beyond the end of a
623 * target, how many sectors must the next target handle?
624 */
625 unsigned short remaining = 0;
626
627 struct dm_target *uninitialized_var(ti);
628 struct queue_limits ti_limits;
629 unsigned i = 0;
630
631 /*
632 * Check each entry in the table in turn.
633 */
634 while (i < dm_table_get_num_targets(table)) {
635 ti = dm_table_get_target(table, i++);
636
637 blk_set_stacking_limits(&ti_limits);
638
639 /* combine all target devices' limits */
640 if (ti->type->iterate_devices)
641 ti->type->iterate_devices(ti, dm_set_device_limits,
642 &ti_limits);
643
644 /*
645 * If the remaining sectors fall entirely within this
646 * table entry are they compatible with its logical_block_size?
647 */
648 if (remaining < ti->len &&
649 remaining & ((ti_limits.logical_block_size >>
650 SECTOR_SHIFT) - 1))
651 break; /* Error */
652
653 next_target_start =
654 (unsigned short) ((next_target_start + ti->len) &
655 (device_logical_block_size_sects - 1));
656 remaining = next_target_start ?
657 device_logical_block_size_sects - next_target_start : 0;
658 }
659
660 if (remaining) {
661 DMWARN("%s: table line %u (start sect %llu len %llu) "
662 "not aligned to h/w logical block size %u",
663 dm_device_name(table->md), i,
664 (unsigned long long) ti->begin,
665 (unsigned long long) ti->len,
666 limits->logical_block_size);
667 return -EINVAL;
668 }
669
670 return 0;
671}
672
673int dm_table_add_target(struct dm_table *t, const char *type,
674 sector_t start, sector_t len, char *params)
675{
676 int r = -EINVAL, argc;
677 char **argv;
678 struct dm_target *tgt;
679
680 if (t->singleton) {
681 DMERR("%s: target type %s must appear alone in table",
682 dm_device_name(t->md), t->targets->type->name);
683 return -EINVAL;
684 }
685
686 BUG_ON(t->num_targets >= t->num_allocated);
687
688 tgt = t->targets + t->num_targets;
689 memset(tgt, 0, sizeof(*tgt));
690
691 if (!len) {
692 DMERR("%s: zero-length target", dm_device_name(t->md));
693 return -EINVAL;
694 }
695
696 tgt->type = dm_get_target_type(type);
697 if (!tgt->type) {
698 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
699 return -EINVAL;
700 }
701
702 if (dm_target_needs_singleton(tgt->type)) {
703 if (t->num_targets) {
704 tgt->error = "singleton target type must appear alone in table";
705 goto bad;
706 }
707 t->singleton = true;
708 }
709
710 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
711 tgt->error = "target type may not be included in a read-only table";
712 goto bad;
713 }
714
715 if (t->immutable_target_type) {
716 if (t->immutable_target_type != tgt->type) {
717 tgt->error = "immutable target type cannot be mixed with other target types";
718 goto bad;
719 }
720 } else if (dm_target_is_immutable(tgt->type)) {
721 if (t->num_targets) {
722 tgt->error = "immutable target type cannot be mixed with other target types";
723 goto bad;
724 }
725 t->immutable_target_type = tgt->type;
726 }
727
728 tgt->table = t;
729 tgt->begin = start;
730 tgt->len = len;
731 tgt->error = "Unknown error";
732
733 /*
734 * Does this target adjoin the previous one ?
735 */
736 if (!adjoin(t, tgt)) {
737 tgt->error = "Gap in table";
738 goto bad;
739 }
740
741 r = dm_split_args(&argc, &argv, params);
742 if (r) {
743 tgt->error = "couldn't split parameters (insufficient memory)";
744 goto bad;
745 }
746
747 r = tgt->type->ctr(tgt, argc, argv);
748 kfree(argv);
749 if (r)
750 goto bad;
751
752 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
753
754 if (!tgt->num_discard_bios && tgt->discards_supported)
755 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
756 dm_device_name(t->md), type);
757
758 return 0;
759
760 bad:
761 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
762 dm_put_target_type(tgt->type);
763 return r;
764}
765
766/*
767 * Target argument parsing helpers.
768 */
769static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
770 unsigned *value, char **error, unsigned grouped)
771{
772 const char *arg_str = dm_shift_arg(arg_set);
773 char dummy;
774
775 if (!arg_str ||
776 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
777 (*value < arg->min) ||
778 (*value > arg->max) ||
779 (grouped && arg_set->argc < *value)) {
780 *error = arg->error;
781 return -EINVAL;
782 }
783
784 return 0;
785}
786
787int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
788 unsigned *value, char **error)
789{
790 return validate_next_arg(arg, arg_set, value, error, 0);
791}
792EXPORT_SYMBOL(dm_read_arg);
793
794int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
795 unsigned *value, char **error)
796{
797 return validate_next_arg(arg, arg_set, value, error, 1);
798}
799EXPORT_SYMBOL(dm_read_arg_group);
800
801const char *dm_shift_arg(struct dm_arg_set *as)
802{
803 char *r;
804
805 if (as->argc) {
806 as->argc--;
807 r = *as->argv;
808 as->argv++;
809 return r;
810 }
811
812 return NULL;
813}
814EXPORT_SYMBOL(dm_shift_arg);
815
816void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
817{
818 BUG_ON(as->argc < num_args);
819 as->argc -= num_args;
820 as->argv += num_args;
821}
822EXPORT_SYMBOL(dm_consume_args);
823
824static bool __table_type_bio_based(unsigned table_type)
825{
826 return (table_type == DM_TYPE_BIO_BASED ||
827 table_type == DM_TYPE_DAX_BIO_BASED);
828}
829
830static bool __table_type_request_based(unsigned table_type)
831{
832 return (table_type == DM_TYPE_REQUEST_BASED ||
833 table_type == DM_TYPE_MQ_REQUEST_BASED);
834}
835
836void dm_table_set_type(struct dm_table *t, unsigned type)
837{
838 t->type = type;
839}
840EXPORT_SYMBOL_GPL(dm_table_set_type);
841
842static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
843 sector_t start, sector_t len, void *data)
844{
845 struct request_queue *q = bdev_get_queue(dev->bdev);
846
847 return q && blk_queue_dax(q);
848}
849
850static bool dm_table_supports_dax(struct dm_table *t)
851{
852 struct dm_target *ti;
853 unsigned i = 0;
854
855 /* Ensure that all targets support DAX. */
856 while (i < dm_table_get_num_targets(t)) {
857 ti = dm_table_get_target(t, i++);
858
859 if (!ti->type->direct_access)
860 return false;
861
862 if (!ti->type->iterate_devices ||
863 !ti->type->iterate_devices(ti, device_supports_dax, NULL))
864 return false;
865 }
866
867 return true;
868}
869
870static int dm_table_determine_type(struct dm_table *t)
871{
872 unsigned i;
873 unsigned bio_based = 0, request_based = 0, hybrid = 0;
874 unsigned sq_count = 0, mq_count = 0;
875 struct dm_target *tgt;
876 struct dm_dev_internal *dd;
877 struct list_head *devices = dm_table_get_devices(t);
878 unsigned live_md_type = dm_get_md_type(t->md);
879
880 if (t->type != DM_TYPE_NONE) {
881 /* target already set the table's type */
882 if (t->type == DM_TYPE_BIO_BASED)
883 return 0;
884 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
885 goto verify_rq_based;
886 }
887
888 for (i = 0; i < t->num_targets; i++) {
889 tgt = t->targets + i;
890 if (dm_target_hybrid(tgt))
891 hybrid = 1;
892 else if (dm_target_request_based(tgt))
893 request_based = 1;
894 else
895 bio_based = 1;
896
897 if (bio_based && request_based) {
898 DMWARN("Inconsistent table: different target types"
899 " can't be mixed up");
900 return -EINVAL;
901 }
902 }
903
904 if (hybrid && !bio_based && !request_based) {
905 /*
906 * The targets can work either way.
907 * Determine the type from the live device.
908 * Default to bio-based if device is new.
909 */
910 if (__table_type_request_based(live_md_type))
911 request_based = 1;
912 else
913 bio_based = 1;
914 }
915
916 if (bio_based) {
917 /* We must use this table as bio-based */
918 t->type = DM_TYPE_BIO_BASED;
919 if (dm_table_supports_dax(t) ||
920 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
921 t->type = DM_TYPE_DAX_BIO_BASED;
922 return 0;
923 }
924
925 BUG_ON(!request_based); /* No targets in this table */
926
927 /*
928 * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
929 * having a compatible target use dm_table_set_type.
930 */
931 t->type = DM_TYPE_REQUEST_BASED;
932
933verify_rq_based:
934 /*
935 * Request-based dm supports only tables that have a single target now.
936 * To support multiple targets, request splitting support is needed,
937 * and that needs lots of changes in the block-layer.
938 * (e.g. request completion process for partial completion.)
939 */
940 if (t->num_targets > 1) {
941 DMWARN("Request-based dm doesn't support multiple targets yet");
942 return -EINVAL;
943 }
944
945 if (list_empty(devices)) {
946 int srcu_idx;
947 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
948
949 /* inherit live table's type and all_blk_mq */
950 if (live_table) {
951 t->type = live_table->type;
952 t->all_blk_mq = live_table->all_blk_mq;
953 }
954 dm_put_live_table(t->md, srcu_idx);
955 return 0;
956 }
957
958 /* Non-request-stackable devices can't be used for request-based dm */
959 list_for_each_entry(dd, devices, list) {
960 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
961
962 if (!blk_queue_stackable(q)) {
963 DMERR("table load rejected: including"
964 " non-request-stackable devices");
965 return -EINVAL;
966 }
967
968 if (q->mq_ops)
969 mq_count++;
970 else
971 sq_count++;
972 }
973 if (sq_count && mq_count) {
974 DMERR("table load rejected: not all devices are blk-mq request-stackable");
975 return -EINVAL;
976 }
977 t->all_blk_mq = mq_count > 0;
978
979 if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) {
980 DMERR("table load rejected: all devices are not blk-mq request-stackable");
981 return -EINVAL;
982 }
983
984 return 0;
985}
986
987unsigned dm_table_get_type(struct dm_table *t)
988{
989 return t->type;
990}
991
992struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
993{
994 return t->immutable_target_type;
995}
996
997struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
998{
999 /* Immutable target is implicitly a singleton */
1000 if (t->num_targets > 1 ||
1001 !dm_target_is_immutable(t->targets[0].type))
1002 return NULL;
1003
1004 return t->targets;
1005}
1006
1007struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1008{
1009 struct dm_target *uninitialized_var(ti);
1010 unsigned i = 0;
1011
1012 while (i < dm_table_get_num_targets(t)) {
1013 ti = dm_table_get_target(t, i++);
1014 if (dm_target_is_wildcard(ti->type))
1015 return ti;
1016 }
1017
1018 return NULL;
1019}
1020
1021bool dm_table_bio_based(struct dm_table *t)
1022{
1023 return __table_type_bio_based(dm_table_get_type(t));
1024}
1025
1026bool dm_table_request_based(struct dm_table *t)
1027{
1028 return __table_type_request_based(dm_table_get_type(t));
1029}
1030
1031bool dm_table_all_blk_mq_devices(struct dm_table *t)
1032{
1033 return t->all_blk_mq;
1034}
1035
1036static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1037{
1038 unsigned type = dm_table_get_type(t);
1039 unsigned per_io_data_size = 0;
1040 struct dm_target *tgt;
1041 unsigned i;
1042
1043 if (unlikely(type == DM_TYPE_NONE)) {
1044 DMWARN("no table type is set, can't allocate mempools");
1045 return -EINVAL;
1046 }
1047
1048 if (__table_type_bio_based(type))
1049 for (i = 0; i < t->num_targets; i++) {
1050 tgt = t->targets + i;
1051 per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1052 }
1053
1054 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1055 if (!t->mempools)
1056 return -ENOMEM;
1057
1058 return 0;
1059}
1060
1061void dm_table_free_md_mempools(struct dm_table *t)
1062{
1063 dm_free_md_mempools(t->mempools);
1064 t->mempools = NULL;
1065}
1066
1067struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1068{
1069 return t->mempools;
1070}
1071
1072static int setup_indexes(struct dm_table *t)
1073{
1074 int i;
1075 unsigned int total = 0;
1076 sector_t *indexes;
1077
1078 /* allocate the space for *all* the indexes */
1079 for (i = t->depth - 2; i >= 0; i--) {
1080 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1081 total += t->counts[i];
1082 }
1083
1084 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1085 if (!indexes)
1086 return -ENOMEM;
1087
1088 /* set up internal nodes, bottom-up */
1089 for (i = t->depth - 2; i >= 0; i--) {
1090 t->index[i] = indexes;
1091 indexes += (KEYS_PER_NODE * t->counts[i]);
1092 setup_btree_index(i, t);
1093 }
1094
1095 return 0;
1096}
1097
1098/*
1099 * Builds the btree to index the map.
1100 */
1101static int dm_table_build_index(struct dm_table *t)
1102{
1103 int r = 0;
1104 unsigned int leaf_nodes;
1105
1106 /* how many indexes will the btree have ? */
1107 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1108 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1109
1110 /* leaf layer has already been set up */
1111 t->counts[t->depth - 1] = leaf_nodes;
1112 t->index[t->depth - 1] = t->highs;
1113
1114 if (t->depth >= 2)
1115 r = setup_indexes(t);
1116
1117 return r;
1118}
1119
1120static bool integrity_profile_exists(struct gendisk *disk)
1121{
1122 return !!blk_get_integrity(disk);
1123}
1124
1125/*
1126 * Get a disk whose integrity profile reflects the table's profile.
1127 * Returns NULL if integrity support was inconsistent or unavailable.
1128 */
1129static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1130{
1131 struct list_head *devices = dm_table_get_devices(t);
1132 struct dm_dev_internal *dd = NULL;
1133 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1134
1135 list_for_each_entry(dd, devices, list) {
1136 template_disk = dd->dm_dev->bdev->bd_disk;
1137 if (!integrity_profile_exists(template_disk))
1138 goto no_integrity;
1139 else if (prev_disk &&
1140 blk_integrity_compare(prev_disk, template_disk) < 0)
1141 goto no_integrity;
1142 prev_disk = template_disk;
1143 }
1144
1145 return template_disk;
1146
1147no_integrity:
1148 if (prev_disk)
1149 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1150 dm_device_name(t->md),
1151 prev_disk->disk_name,
1152 template_disk->disk_name);
1153 return NULL;
1154}
1155
1156/*
1157 * Register the mapped device for blk_integrity support if the
1158 * underlying devices have an integrity profile. But all devices may
1159 * not have matching profiles (checking all devices isn't reliable
1160 * during table load because this table may use other DM device(s) which
1161 * must be resumed before they will have an initialized integity
1162 * profile). Consequently, stacked DM devices force a 2 stage integrity
1163 * profile validation: First pass during table load, final pass during
1164 * resume.
1165 */
1166static int dm_table_register_integrity(struct dm_table *t)
1167{
1168 struct mapped_device *md = t->md;
1169 struct gendisk *template_disk = NULL;
1170
1171 template_disk = dm_table_get_integrity_disk(t);
1172 if (!template_disk)
1173 return 0;
1174
1175 if (!integrity_profile_exists(dm_disk(md))) {
1176 t->integrity_supported = true;
1177 /*
1178 * Register integrity profile during table load; we can do
1179 * this because the final profile must match during resume.
1180 */
1181 blk_integrity_register(dm_disk(md),
1182 blk_get_integrity(template_disk));
1183 return 0;
1184 }
1185
1186 /*
1187 * If DM device already has an initialized integrity
1188 * profile the new profile should not conflict.
1189 */
1190 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1191 DMWARN("%s: conflict with existing integrity profile: "
1192 "%s profile mismatch",
1193 dm_device_name(t->md),
1194 template_disk->disk_name);
1195 return 1;
1196 }
1197
1198 /* Preserve existing integrity profile */
1199 t->integrity_supported = true;
1200 return 0;
1201}
1202
1203/*
1204 * Prepares the table for use by building the indices,
1205 * setting the type, and allocating mempools.
1206 */
1207int dm_table_complete(struct dm_table *t)
1208{
1209 int r;
1210
1211 r = dm_table_determine_type(t);
1212 if (r) {
1213 DMERR("unable to determine table type");
1214 return r;
1215 }
1216
1217 r = dm_table_build_index(t);
1218 if (r) {
1219 DMERR("unable to build btrees");
1220 return r;
1221 }
1222
1223 r = dm_table_register_integrity(t);
1224 if (r) {
1225 DMERR("could not register integrity profile.");
1226 return r;
1227 }
1228
1229 r = dm_table_alloc_md_mempools(t, t->md);
1230 if (r)
1231 DMERR("unable to allocate mempools");
1232
1233 return r;
1234}
1235
1236static DEFINE_MUTEX(_event_lock);
1237void dm_table_event_callback(struct dm_table *t,
1238 void (*fn)(void *), void *context)
1239{
1240 mutex_lock(&_event_lock);
1241 t->event_fn = fn;
1242 t->event_context = context;
1243 mutex_unlock(&_event_lock);
1244}
1245
1246void dm_table_event(struct dm_table *t)
1247{
1248 /*
1249 * You can no longer call dm_table_event() from interrupt
1250 * context, use a bottom half instead.
1251 */
1252 BUG_ON(in_interrupt());
1253
1254 mutex_lock(&_event_lock);
1255 if (t->event_fn)
1256 t->event_fn(t->event_context);
1257 mutex_unlock(&_event_lock);
1258}
1259EXPORT_SYMBOL(dm_table_event);
1260
1261sector_t dm_table_get_size(struct dm_table *t)
1262{
1263 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1264}
1265EXPORT_SYMBOL(dm_table_get_size);
1266
1267struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1268{
1269 if (index >= t->num_targets)
1270 return NULL;
1271
1272 return t->targets + index;
1273}
1274
1275/*
1276 * Search the btree for the correct target.
1277 *
1278 * Caller should check returned pointer with dm_target_is_valid()
1279 * to trap I/O beyond end of device.
1280 */
1281struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1282{
1283 unsigned int l, n = 0, k = 0;
1284 sector_t *node;
1285
1286 for (l = 0; l < t->depth; l++) {
1287 n = get_child(n, k);
1288 node = get_node(t, l, n);
1289
1290 for (k = 0; k < KEYS_PER_NODE; k++)
1291 if (node[k] >= sector)
1292 break;
1293 }
1294
1295 return &t->targets[(KEYS_PER_NODE * n) + k];
1296}
1297
1298static int count_device(struct dm_target *ti, struct dm_dev *dev,
1299 sector_t start, sector_t len, void *data)
1300{
1301 unsigned *num_devices = data;
1302
1303 (*num_devices)++;
1304
1305 return 0;
1306}
1307
1308/*
1309 * Check whether a table has no data devices attached using each
1310 * target's iterate_devices method.
1311 * Returns false if the result is unknown because a target doesn't
1312 * support iterate_devices.
1313 */
1314bool dm_table_has_no_data_devices(struct dm_table *table)
1315{
1316 struct dm_target *uninitialized_var(ti);
1317 unsigned i = 0, num_devices = 0;
1318
1319 while (i < dm_table_get_num_targets(table)) {
1320 ti = dm_table_get_target(table, i++);
1321
1322 if (!ti->type->iterate_devices)
1323 return false;
1324
1325 ti->type->iterate_devices(ti, count_device, &num_devices);
1326 if (num_devices)
1327 return false;
1328 }
1329
1330 return true;
1331}
1332
1333/*
1334 * Establish the new table's queue_limits and validate them.
1335 */
1336int dm_calculate_queue_limits(struct dm_table *table,
1337 struct queue_limits *limits)
1338{
1339 struct dm_target *uninitialized_var(ti);
1340 struct queue_limits ti_limits;
1341 unsigned i = 0;
1342
1343 blk_set_stacking_limits(limits);
1344
1345 while (i < dm_table_get_num_targets(table)) {
1346 blk_set_stacking_limits(&ti_limits);
1347
1348 ti = dm_table_get_target(table, i++);
1349
1350 if (!ti->type->iterate_devices)
1351 goto combine_limits;
1352
1353 /*
1354 * Combine queue limits of all the devices this target uses.
1355 */
1356 ti->type->iterate_devices(ti, dm_set_device_limits,
1357 &ti_limits);
1358
1359 /* Set I/O hints portion of queue limits */
1360 if (ti->type->io_hints)
1361 ti->type->io_hints(ti, &ti_limits);
1362
1363 /*
1364 * Check each device area is consistent with the target's
1365 * overall queue limits.
1366 */
1367 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1368 &ti_limits))
1369 return -EINVAL;
1370
1371combine_limits:
1372 /*
1373 * Merge this target's queue limits into the overall limits
1374 * for the table.
1375 */
1376 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1377 DMWARN("%s: adding target device "
1378 "(start sect %llu len %llu) "
1379 "caused an alignment inconsistency",
1380 dm_device_name(table->md),
1381 (unsigned long long) ti->begin,
1382 (unsigned long long) ti->len);
1383 }
1384
1385 return validate_hardware_logical_block_alignment(table, limits);
1386}
1387
1388/*
1389 * Verify that all devices have an integrity profile that matches the
1390 * DM device's registered integrity profile. If the profiles don't
1391 * match then unregister the DM device's integrity profile.
1392 */
1393static void dm_table_verify_integrity(struct dm_table *t)
1394{
1395 struct gendisk *template_disk = NULL;
1396
1397 if (t->integrity_supported) {
1398 /*
1399 * Verify that the original integrity profile
1400 * matches all the devices in this table.
1401 */
1402 template_disk = dm_table_get_integrity_disk(t);
1403 if (template_disk &&
1404 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1405 return;
1406 }
1407
1408 if (integrity_profile_exists(dm_disk(t->md))) {
1409 DMWARN("%s: unable to establish an integrity profile",
1410 dm_device_name(t->md));
1411 blk_integrity_unregister(dm_disk(t->md));
1412 }
1413}
1414
1415static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1416 sector_t start, sector_t len, void *data)
1417{
1418 unsigned long flush = (unsigned long) data;
1419 struct request_queue *q = bdev_get_queue(dev->bdev);
1420
1421 return q && (q->queue_flags & flush);
1422}
1423
1424static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1425{
1426 struct dm_target *ti;
1427 unsigned i = 0;
1428
1429 /*
1430 * Require at least one underlying device to support flushes.
1431 * t->devices includes internal dm devices such as mirror logs
1432 * so we need to use iterate_devices here, which targets
1433 * supporting flushes must provide.
1434 */
1435 while (i < dm_table_get_num_targets(t)) {
1436 ti = dm_table_get_target(t, i++);
1437
1438 if (!ti->num_flush_bios)
1439 continue;
1440
1441 if (ti->flush_supported)
1442 return true;
1443
1444 if (ti->type->iterate_devices &&
1445 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1446 return true;
1447 }
1448
1449 return false;
1450}
1451
1452static bool dm_table_discard_zeroes_data(struct dm_table *t)
1453{
1454 struct dm_target *ti;
1455 unsigned i = 0;
1456
1457 /* Ensure that all targets supports discard_zeroes_data. */
1458 while (i < dm_table_get_num_targets(t)) {
1459 ti = dm_table_get_target(t, i++);
1460
1461 if (ti->discard_zeroes_data_unsupported)
1462 return false;
1463 }
1464
1465 return true;
1466}
1467
1468static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1469 sector_t start, sector_t len, void *data)
1470{
1471 struct request_queue *q = bdev_get_queue(dev->bdev);
1472
1473 return q && blk_queue_nonrot(q);
1474}
1475
1476static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1477 sector_t start, sector_t len, void *data)
1478{
1479 struct request_queue *q = bdev_get_queue(dev->bdev);
1480
1481 return q && !blk_queue_add_random(q);
1482}
1483
1484static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1485 sector_t start, sector_t len, void *data)
1486{
1487 struct request_queue *q = bdev_get_queue(dev->bdev);
1488
1489 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1490}
1491
1492static bool dm_table_all_devices_attribute(struct dm_table *t,
1493 iterate_devices_callout_fn func)
1494{
1495 struct dm_target *ti;
1496 unsigned i = 0;
1497
1498 while (i < dm_table_get_num_targets(t)) {
1499 ti = dm_table_get_target(t, i++);
1500
1501 if (!ti->type->iterate_devices ||
1502 !ti->type->iterate_devices(ti, func, NULL))
1503 return false;
1504 }
1505
1506 return true;
1507}
1508
1509static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1510 sector_t start, sector_t len, void *data)
1511{
1512 struct request_queue *q = bdev_get_queue(dev->bdev);
1513
1514 return q && !q->limits.max_write_same_sectors;
1515}
1516
1517static bool dm_table_supports_write_same(struct dm_table *t)
1518{
1519 struct dm_target *ti;
1520 unsigned i = 0;
1521
1522 while (i < dm_table_get_num_targets(t)) {
1523 ti = dm_table_get_target(t, i++);
1524
1525 if (!ti->num_write_same_bios)
1526 return false;
1527
1528 if (!ti->type->iterate_devices ||
1529 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1530 return false;
1531 }
1532
1533 return true;
1534}
1535
1536static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1537 sector_t start, sector_t len, void *data)
1538{
1539 struct request_queue *q = bdev_get_queue(dev->bdev);
1540
1541 return q && blk_queue_discard(q);
1542}
1543
1544static bool dm_table_supports_discards(struct dm_table *t)
1545{
1546 struct dm_target *ti;
1547 unsigned i = 0;
1548
1549 /*
1550 * Unless any target used by the table set discards_supported,
1551 * require at least one underlying device to support discards.
1552 * t->devices includes internal dm devices such as mirror logs
1553 * so we need to use iterate_devices here, which targets
1554 * supporting discard selectively must provide.
1555 */
1556 while (i < dm_table_get_num_targets(t)) {
1557 ti = dm_table_get_target(t, i++);
1558
1559 if (!ti->num_discard_bios)
1560 continue;
1561
1562 if (ti->discards_supported)
1563 return true;
1564
1565 if (ti->type->iterate_devices &&
1566 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1567 return true;
1568 }
1569
1570 return false;
1571}
1572
1573void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1574 struct queue_limits *limits)
1575{
1576 bool wc = false, fua = false;
1577
1578 /*
1579 * Copy table's limits to the DM device's request_queue
1580 */
1581 q->limits = *limits;
1582
1583 if (!dm_table_supports_discards(t))
1584 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1585 else
1586 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1587
1588 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1589 wc = true;
1590 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1591 fua = true;
1592 }
1593 blk_queue_write_cache(q, wc, fua);
1594
1595 if (!dm_table_discard_zeroes_data(t))
1596 q->limits.discard_zeroes_data = 0;
1597
1598 /* Ensure that all underlying devices are non-rotational. */
1599 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1600 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1601 else
1602 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1603
1604 if (!dm_table_supports_write_same(t))
1605 q->limits.max_write_same_sectors = 0;
1606
1607 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1608 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1609 else
1610 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1611
1612 dm_table_verify_integrity(t);
1613
1614 /*
1615 * Determine whether or not this queue's I/O timings contribute
1616 * to the entropy pool, Only request-based targets use this.
1617 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1618 * have it set.
1619 */
1620 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1621 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1622
1623 /*
1624 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1625 * visible to other CPUs because, once the flag is set, incoming bios
1626 * are processed by request-based dm, which refers to the queue
1627 * settings.
1628 * Until the flag set, bios are passed to bio-based dm and queued to
1629 * md->deferred where queue settings are not needed yet.
1630 * Those bios are passed to request-based dm at the resume time.
1631 */
1632 smp_mb();
1633 if (dm_table_request_based(t))
1634 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1635}
1636
1637unsigned int dm_table_get_num_targets(struct dm_table *t)
1638{
1639 return t->num_targets;
1640}
1641
1642struct list_head *dm_table_get_devices(struct dm_table *t)
1643{
1644 return &t->devices;
1645}
1646
1647fmode_t dm_table_get_mode(struct dm_table *t)
1648{
1649 return t->mode;
1650}
1651EXPORT_SYMBOL(dm_table_get_mode);
1652
1653enum suspend_mode {
1654 PRESUSPEND,
1655 PRESUSPEND_UNDO,
1656 POSTSUSPEND,
1657};
1658
1659static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1660{
1661 int i = t->num_targets;
1662 struct dm_target *ti = t->targets;
1663
1664 while (i--) {
1665 switch (mode) {
1666 case PRESUSPEND:
1667 if (ti->type->presuspend)
1668 ti->type->presuspend(ti);
1669 break;
1670 case PRESUSPEND_UNDO:
1671 if (ti->type->presuspend_undo)
1672 ti->type->presuspend_undo(ti);
1673 break;
1674 case POSTSUSPEND:
1675 if (ti->type->postsuspend)
1676 ti->type->postsuspend(ti);
1677 break;
1678 }
1679 ti++;
1680 }
1681}
1682
1683void dm_table_presuspend_targets(struct dm_table *t)
1684{
1685 if (!t)
1686 return;
1687
1688 suspend_targets(t, PRESUSPEND);
1689}
1690
1691void dm_table_presuspend_undo_targets(struct dm_table *t)
1692{
1693 if (!t)
1694 return;
1695
1696 suspend_targets(t, PRESUSPEND_UNDO);
1697}
1698
1699void dm_table_postsuspend_targets(struct dm_table *t)
1700{
1701 if (!t)
1702 return;
1703
1704 suspend_targets(t, POSTSUSPEND);
1705}
1706
1707int dm_table_resume_targets(struct dm_table *t)
1708{
1709 int i, r = 0;
1710
1711 for (i = 0; i < t->num_targets; i++) {
1712 struct dm_target *ti = t->targets + i;
1713
1714 if (!ti->type->preresume)
1715 continue;
1716
1717 r = ti->type->preresume(ti);
1718 if (r) {
1719 DMERR("%s: %s: preresume failed, error = %d",
1720 dm_device_name(t->md), ti->type->name, r);
1721 return r;
1722 }
1723 }
1724
1725 for (i = 0; i < t->num_targets; i++) {
1726 struct dm_target *ti = t->targets + i;
1727
1728 if (ti->type->resume)
1729 ti->type->resume(ti);
1730 }
1731
1732 return 0;
1733}
1734
1735void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1736{
1737 list_add(&cb->list, &t->target_callbacks);
1738}
1739EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1740
1741int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1742{
1743 struct dm_dev_internal *dd;
1744 struct list_head *devices = dm_table_get_devices(t);
1745 struct dm_target_callbacks *cb;
1746 int r = 0;
1747
1748 list_for_each_entry(dd, devices, list) {
1749 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1750 char b[BDEVNAME_SIZE];
1751
1752 if (likely(q))
1753 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1754 else
1755 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1756 dm_device_name(t->md),
1757 bdevname(dd->dm_dev->bdev, b));
1758 }
1759
1760 list_for_each_entry(cb, &t->target_callbacks, list)
1761 if (cb->congested_fn)
1762 r |= cb->congested_fn(cb, bdi_bits);
1763
1764 return r;
1765}
1766
1767struct mapped_device *dm_table_get_md(struct dm_table *t)
1768{
1769 return t->md;
1770}
1771EXPORT_SYMBOL(dm_table_get_md);
1772
1773void dm_table_run_md_queue_async(struct dm_table *t)
1774{
1775 struct mapped_device *md;
1776 struct request_queue *queue;
1777 unsigned long flags;
1778
1779 if (!dm_table_request_based(t))
1780 return;
1781
1782 md = dm_table_get_md(t);
1783 queue = dm_get_md_queue(md);
1784 if (queue) {
1785 if (queue->mq_ops)
1786 blk_mq_run_hw_queues(queue, true);
1787 else {
1788 spin_lock_irqsave(queue->queue_lock, flags);
1789 blk_run_queue_async(queue);
1790 spin_unlock_irqrestore(queue->queue_lock, flags);
1791 }
1792 }
1793}
1794EXPORT_SYMBOL(dm_table_run_md_queue_async);
1795